OBSCURATION IN EXTREMELY LUMINOUS QUASARS

The Astrophysical Journal, 675:960Y984, 2008 March 10 # 2008. The American Astronomical Society. All rights reserved. Printed in U.S.A. A OBSCURATION IN EXTREMELY LUMINOUS QUASARS M. Polletta, 1,2 D. Weedman, 3 S. Hönig, 4 C. J. Lonsdale, 1,5 H. E. Smith, 1,6 and J. Houck 3 Received 2007 April 11; accepted 2007 September 26 ABSTRACT The SEDs and IR spectra of a remarkable sample of obscured AGNs selected in the MIR are modeled with recent clumpy torus models. The sample contains 21 AGNs at z ¼ 1:3Y3 discovered in the largest Spitzer surveys (SWIRE, NDWFS, and FLS) by means of their extremely red IR to optical colors. All sources show the 9.7 m silicate feature in absorption and have extreme MIR luminosities [L(6 m) 10 46 ergs s 1 ]. The IR SEDs and spectra of 12 sources are well reproduced with a simple torus model, while the remaining nine sources require foreground extinction from a cold dust component to reproduce both the depth of the silicate feature and the NIR emission from hot dust. The best-fit torus models show a broad range of inclinations. Based on the unobscured QSO MIR luminosity function ( Brown and coworkers) and on a color-selected sample of AGNs, we estimate the surface densities of obscured and unobscured QSOs with L(6 m) > 10 12 L and z ¼ 1:3Y3:0 to be about 17Y22 and 11.7 deg 2, respectively. Overall we find that 35%Y41% of luminous QSOs are unobscured, 37%Y40% are obscured by the torus, and 23%Y25% are obscured by a cold absorber detached from the torus. These fractions are consistent with a decrease of the torus covering fraction at large luminosities as predicted by receding torus models. An FIR component is observed in eight objects with luminosity greater than 3:3 ; 10 12 L, implying SFRs of 600Y3000 M yr 1. In the whole sample, the average contribution from a starburst component to the bolometric luminosity, as estimated from the PAH 7.7 m luminosity in the composite IR spectra, is 20% of the total bolometric luminosity. Subject headinggs: galaxies: active galaxies: high-redshift infrared: galaxies quasars: general Online material: color figures 1 Center for Astrophysics and Space Sciences, University of California, San Diego, La Jolla, CA 92093-0424. 2 Institut d Astrophysique de Paris, 75014 Paris, France; polletta@iap.fr. 3 Department of Astronomy, Cornell University, Ithaca, NY 14853. 4 Max-Planck-Institut für Radioastronomie, 53121 Bonn, Germany. 5 Infrared Processing and Analysis Center, California Institute of Technology, Pasadena, CA 91125. 6 Deceased 2007 August 16. 960 1. INTRODUCTION 1.1. Obscured AGNs at High Luminosities Multiple X-ray studies show that the fraction of active galactic nuclei (AGNs) whose emission is heavily absorbed (N H 10 22 cm 2 ) decreases with increasing luminosity (from >80% at L X ¼ 10 42 ergs s 1 to 38% at 10 45 ergs s 1 ; Akylas et al. 2006). However, because of the difficulty in detecting and identifying absorbed AGNs, it is still unclear whether their paucity at high luminosities is an observational selection effect or real ( La Franca et al. 2005; Treister & Urry 2006; Akylas et al. 2006; Tozzi et al. 2006). In order to overcome these uncertainties, searches for absorbed QSOs have focused on observations at wavelengths less affected by absorption, i.e., infrared ( IR) and radio, e.g., FIRST, 2MASS, and various Spitzer surveys (e.g., Polletta et al. 2006; Martínez-Sansigre et al. 2006; Lacy et al. 2007; Wilkes et al. 2002; Urrutia et al. 2005). These searches have unveiled a large population of QSOs obscured at optical wavelengths. Assuming that all optically obscured QSOsarealsoabsorbedinX-rays,thefractionofabsorbedQSOs would be 50% of all QSOs (Martínez-Sansigre et al. 2005). This fraction is still significantly lower than the fraction measured for AGNs at lower luminosities, 80% (Osterbrock & Martel 1993; Akylas et al. 2006). Current AGN evolutionary models and observations indicate a link between the absorption properties in AGNs and their luminosities (see, e.g., Fig. 4 in Hopkins et al. 2005). More specifically, AGNs in a growing phase are moderately luminous and heavily absorbed, and as the AGN luminosity increases, the intense radiation field destroys the surrounding matter and the AGN shines unabsorbed (e.g., Di Matteo et al. 2005). A similar trend is predicted by the receding torus models (e.g., Lawrence 1991; Simpson 2005). According to these models, the opening angle of the torus (measured from the torus axis to the equatorial plane) is larger in more luminous objects. Thus, obscuration should be less common in more luminous AGNs. 1.2. The Obscuring Matter in AGNs Obscuration in AGNs is caused by a mixture of neutral and ionized gas, as well as dust. Absorption by gas is usually observed in the X-rays through the suppression of soft X-ray emission with an energy cutoff that increases with the gas column density, N H. The absorbing gas is mainly located in the circumnuclear region as suggested by measured variability ( Risaliti et al. 2005, 2007), but gas in our Galaxy and in the host galaxy can also contribute to the overall absorption. Obscuration by dust is usually observed at optical and IR wavelengths where the continuum and broad emission lines are reddened and, in some cases, completely suppressed. In the near- and mid-ir (NIR and MIR), dust obscuration interplays with thermal reemission from the putative parsec-scaled dust torus. Depending on the actual geometry, optical depth, and chemistry, dust absorption features at about 10 and 18 m due to silicates (Si) can be present in the IR. According to AGN unification models (e.g., Antonucci 1993) and torus models (e.g., Krolik & Begelman 1988; Pier & Krolik 1992; Efstathiou & Rowan-Robinson 1995; Granato et al. 1997; Nenkova et al. 2002; Dullemond & van Bemmel 2005; Fritz et al. 2006; Hönig et al. 2006), optical depths are expected to be at least roughly correlated from the X-ray through the MIR, although this correlation might be very weak for clumpy tori. In obscured

OBSCURATION IN EXTREMELY LUMINOUS QUASARS 961 AGNs the line of sight (LOS) intercepts the putative torus and thus absorbs radiation from the nucleus ( broad emission lines, X-rays, and thermal radiation from hot dust), while in unobscured AGNs the radiation from the nucleus is directly visible and thus unabsorbed. An edge-on view of the torus is characterized by red optical through IR colors and Si features in absorption (at 10 and 18 m), while a face-on torus has bluer colors and Si features absent or in emission. Sources with narrow emission lines in their optical spectra, red colors, and strong Si absorption features are usually associated with a highly inclined torus and thus are expected to be also absorbed in the X-rays. However, there is emerging evidence for a surprising mismatch between the absorption measured in the X-rays and that measured in the optical through IR for a significant fraction of sources, especially at high luminosities. AGNs selected in the MIR because of extremely red optical through IR colors sometimes show broad optical emission lines, even having absorption features in their IR spectra in some cases ( Brand et al. 2007). The comparison between X-ray absorption and optical obscuration in various X-rayYselected AGN samples shows that about 20%Y30% of obscured AGNs are not absorbed in the X-rays and vice versa (e.g., Perola et al. 2004; Tozzi et al. 2006; Tajer et al. 2007; Gliozzi et al. 2007). The MIR spectra of a sample of type 2 QSOs with heavy X-ray absorption did not reveal the Si feature in absorption at 10 m (Sturm et al. 2006) as expected. This mismatch is even more pronounced at high luminosities. Indeed, absorption signatures in the IR, e.g., the Si absorption feature at 10 m, are more prominent at the high luminosities of ultraluminous infrared galaxies ( ULIRGS; Sanders & Mirabel 1996; Hao et al. 2007), while X-ray absorption is progressively less common in AGNs with increasing luminosity (Ueda et al. 2003; Hasinger et al. 2005). These results suggest that a nonnegligible fraction of obscured AGNs might not be obscured by a torus, but by dust in either the narrow-line region (NLR; Sturm et al. 2005) or the host galaxy, and that large X-ray column densities are not always associated with geometrically and optically thick dust distributions (see also Rigby et al. 2006; Martínez-Sansigre et al. 2006). Alternatively, this mismatch can be explained by a low dust-to-gas ratio (A V /N H ) or by a different path for the IR and the X-ray LOSs (Maiolino et al. 2001; Shi et al. 2006). Large optical depths for Si absorption features (>1.7) usually imply a compact source deeply embedded in a smooth distribution of material that is both geometrically and optically thick, rather than absorption by a foreground screen of dust close to the heating source (Levenson et al. 2007; Imanishi et al. 2007). Also a detached foreground cold absorber that is far enough from the heating source to be cold can produce large Si optical depths. Deep ( Si > 1:7) Si absorption features are usually considered as indicators of a buried compact source such as an AGN. However, it is not clear whether the optical depth is a consequence of a random alignment of Si clouds or a specific dust distribution and orientation. The observed MIR spectra of obscured AGNs do not favor the latter scenario because the Si absorption features are not always present in the spectra of optically obscured and X-ray absorbed AGNs, as would be expected if the Si absorbers were associated with the same material that suppresses the broad emission lines and the soft X-ray emission. In order to investigate the properties of the obscuring matter in extremely luminous AGNs and asses how often the absorption signatures at optical, X-ray, and infrared wavelengths do not correlate, a comprehensive analysis of the absorption properties at all wavelengths of obscured AGNs at high luminosity is necessary, as well as detailed studies of their absorbing matter. Observations at various wavelengths from X-ray through optical and IR constrain the absorption along multiple LOSs and thus the geometry and distribution of the absorbing matter. In this work, we investigate the observed properties of the obscuring matter in a sample of obscured and extremely luminous AGNs, and we model their spectral energy distributions (SEDs) with clumpy torus models (Hönig et al. 2006) to explore the dust geometries associated with large obscuration. This study is based on a sample of extremely luminous and obscured AGNs for which multiwavelength SEDs and IR spectra are available. The selected sample includes sources from the three widest Spitzer extragalactic surveys that were observed with the Spitzer Infrared Spectrograph (IRS; Houck et al. 2004). In x 2 we describe the sample selection, and in x 3 we present the observations and the data used in this work. The IRS spectra and detected spectral features are presented in x 4. The modeling of the SEDs using clumpy torusmodels(hönig et al. 2006) is presented in x 5. The composite spectra of two subsamples defined on the basis of the model results are analyzed in x 6. The properties (IR colors, surface density, luminosity, optical depths, and redshift distribution) of the selected sample are compared with those of other samples of AGNs and ULIRGs from the literature in x 7. The X-ray properties and a comparisonbetweentheabsorptionseeninthex-rayandtheoptical depth in the IR are described in x 8. Our results and their implications are discussed in x 9 and summarized in x 10. Throughout the paper, the term absorbed refers to X-ray sources with effective column densities N H 10 22 cm 2, and obscured to sources with red opticalyir colors, e.g., F(3:6 m)/ F(r 0 ) > 20 or F(24 m)/f(r 0 ) > 100, implying extreme dust extinction. The IR SEDs are defined as red if they are as red or redder than a power-law model, F /, with slope ¼ 2. The terms type 1 and type 2 refer to AGNs with broad and narrow optical emission lines, respectively, in their optical spectra. We adopt a flat cosmology with H 0 ¼ 71 km s 1 Mpc 1, M ¼ 0:27, and ¼ 0:73 (Spergel et al. 2003). 2. SAMPLE DESCRIPTION There are three large-area Spitzer surveys with the Infrared Array Camera ( IRAC; Fazio et al. 2004), the Multiband Imaging Photometer ( MIPS; Rieke et al. 2004), and optical photometric coverage, the 50 deg 2 Spitzer Wide-Area Infrared Extragalactic Survey 7 (SWIRE; Lonsdale et al. 2003, 2004); the 9 deg 2 NOAO Deep Wide-Field Survey ( NDWFS) of the Bootes field (Jannuzi & Dey1999; Murray et al. 2005); and the 3.7 deg 2 Extragalactic First Look Survey 8 (E-FLS; Fadda et al. 2006). The initial objective for Spitzer IRS spectroscopy of sources chosen from these surveys was to understand populations of new sources that would not have been known prior to Spitzer. Consequently, spectroscopy has emphasized sources that are optically very faint, typically mag(r) > 24, and IR-bright, i.e., F(24 m) > 1 mjy. Various selection criteria based on IRAC colors also entered choices of spectroscopic targets, but the unifying theme of existing spectroscopic samples is their high ratio of MIR to optical flux, IR/opt ¼ F (24 m)/f (R). Most sources chosen for spectroscopy have IR/opt 10. For 58 sources observed in Bootes (Houck et al. 2005; Weedman et al. 2006c), 70 sources observed in the FLS (Yan et al. 2007; Weedman et al. 2006b), and 20 sources in the SWIRE Lockman Hole field (Weedman et al. 2006a), the majority are at high redshift (z 2) and many have infrared spectra dominated by the strong silicate (Si) absorption feature at rest frame 9.7 m. The presence of this absorption combined with 7 See http://swire.ipac.caltech.edu /swire/swire.html. 8 See http://ssc.spitzer.caltech.edu/fls/.

962 POLLETTA ET AL. the red colors and high luminosities of these sources led these previous observers to the conclusion that IR-bright sources with large IR/optical flux ratios are dominated by obscured AGNs (or QSOs because of their very high luminosities). A minority of sources in these studies show strong PAH emission features characteristic of starbursts, but these sources generally do not show strong Si absorption, so different populations of AGNs and starbursts are observed. If the interpretation of the obscured sources as AGNs is correct, this implies a significant population of luminous, obscured QSOs. It is essential, therefore, to verify the AGN classification and to understand the obscured sources in the context of models for AGNs. Our goal in this work is to determine quantitative AGN models for the most luminous of these obscured sources, with two primary objectives. The first is to investigate the properties of the AGN emission and the nature of the obscuration in AGNs at high luminosities. The second is to use the SEDs to define the color criteria that describe such obscured AGNs in the IRAC and MIPS color space in order to compare surface densities between obscured and unobscured QSOs at high luminosities and high redshifts. To achieve these goals, we have selected for detailed modeling the most luminous of the obscured sources with AGN-dominated MIR SEDs from the archival Spitzer spectroscopy in the survey fields mentioned. Our selection criteria are the following: high redshift, i.e., z > 1; large MIR luminosities, i.e., 6 m luminosities, L (6 m) ¼ L(6 m) 10 12 L ; the presence of Si in absorption in the IRS spectra; and red MIR SEDs consistent with being AGN dominated (see, e.g., Polletta et al. 2006; Alonso-Herrero et al. 2006). There are 21 sources that satisfy all these criteria; 13 are from the literature (Houck et al. 2005; Weedman et al. 2006a; Yan et al. 2007), and 8 are from our own IRS observing programs. As a consequence of these selection criteria, all sources have 24 m fluxes greater than 1 mjy and have relatively large IR/optical flux ratios, i.e., F(24 m)/f(r 0 ) 400. To our knowledge, this sample contains all sources with available IRS spectra in the literature or the Spitzer public archive that satisfy the selection criteria described. However, this sample is not meant to be complete or unbiased since it is mainly based on the availability of IRS spectra, therefore it is not essential to be exhaustive. The selected sample includes the most luminous obscured QSOs currently known. The large MIR luminosities of the selected targets enable us to measure the emission from the hottest dust in the torus with negligible contribution from the host galaxy, providing a laboratory to test the predictions of torus models in obscured AGNs. Up to now, these kinds of SEDs could be sampled only using high spatial resolution data in nearby AGNs (e.g., in NGC 1068; Hönig et al. 2006). By combining samples from the widest Spitzer surveys, we were able to find a significant number of these rare objects. The list of selected sources, coordinates, and optical and IR fluxes are reported in Table 1. For simplicity, we use throughout the paper simplified names for the selected sources. Official IAU names are reported in Table 1. For the SWIRE sources, the first two letters in their names designate the field where they were discovered (LH for Lockman Hole, N1 for ELAIS-N1, and N2 for ELAIS-N2). The first letter of the NDWFS source names is B for Bootes. The names of the FLS sources correspond to those in Yan et al. (2007). 3. OBSERVATIONS AND DATA ANALYSIS Multiband photometric data in the optical from the ground and in the IR from Spitzer are available for all sources. The available data and references are listed in Table 1. All sources are detected with MIPS at 24 m with fluxes ranging from 1.0 to 10.6 mjy and a median flux of 2.6 mjy. All sources have been observed with IRAC in four bands, from 3.6 to 8.0 m with varying limits. The FLS sources have few detections with IRAC, probably because of their less sensitive observations compared to the NDWFS and SWIRE surveys. Eight sources have been observed in five optical bands, from U to z band, three in four bands, five in three bands, and five in one band. Optical magnitudes for Bootes sources are in NDWFS Data Release 3 (DR3). Optical and IR fluxes for the FLS sources were obtained from Yan et al. (2007) and Sajina et al. (2007a), and those for the SWIRE sources were taken from the latest internal catalogs that will be released in the Data Release 6 (for details on the data reduction see Surace et al. 2005). 9 The IRAC fluxes of the sources in the NDWFS were measured from the post-bcd images in the Spitzer archive. Aperture fluxes were measured within a 2.9 00 radius aperture at their position and background subtracted. Aperture corrections as derived from the SWIRE survey were applied. The aperture corrections are 1.15, 1.15, 1.25, and 1.43 for the four IRAC bands, respectively. All sources have been observed in the optical r 0 or R band, and 16/21 have been detected. The F(24 m)/f(r) flux ratio ranges from 575 to 25,000 with a median value of 3500. IRS data were taken in part from the archive and in part from our own projects (Houck et al. 2005; Weedman et al. 2006a). The spectroscopic observations were made with the IRS Short Low module in order 1 (SL1) and with the Long Low module inorders1and2(ll1andll2),describedinhoucketal. (2004), 10 providing low-resolution spectral coverage from 8to 35 m. The main parameters of the IRS observations are listed in Table 2. Since the objects have similar properties in terms of SED and fluxes, we applied the same reduction procedure to all spectra for uniformity. The reduction method is described in detail in Weedman et al. (2006a) and briefly summarized here. Six SWIRE sources (see details in Table 2) were processed with version 11.0 of the SSC pipeline; the remaining sources were processed with version 13.0. Extraction of source spectra was done with the SMART analysis package ( Higdon et al. 2004). Since the selected sources are faint and the spectra are dominated by background signal, we restricted the number of pixels used to define the source spectrum to a width of only 4 pixels (scaling with wavelength) in order to improve the spectral signal-to-noise ratio (S/ N). The flux correction necessary to change the fluxes obtained with the narrow extraction to the fluxes that would be measured with the standard extraction provided for the basic calibrated data was measured by extracting an unresolved source of high S/N with both techniques, and this correction (typically 10% but varying with wavelength) was applied to all sources. A search in the literature provided additional data for six sources in the sample, an Infrared Space Observatory (ISO) 15m flux measurement for N2_08 (González-Solares et al. 2005), and MAMBO 1.2 mm flux upper limits for all five E-FLS sources (Lutz et al. 2005). The SEDs and IRS spectra of all sources in the sample are shown in Figures 1 and 2. The displayed spectra have been boxcar smoothed to 9 times the approximate resolution of the different IRS modules (0.6 m for SL1, 0.9 m for LL2, and 1.2 m for 9 Available at http://data.spitzer.caltech.edu/popular/swire/20050603_ enhanced_v1/ Documents/ SWIRE2_doc_083105.pdf. 10 The IRS was a collaborative venture between Cornell University and Ball Aerospace Corporation funded by NASA through the Jet Propulsion Laboratory and the Ames Research Center.

TABLE 1 Optical and Infrared Data of the Obscured QSO Sample Survey Name Source Name IAU Name F U a (Jy) Fg 0 or B a ( Jy) Fr 0 or R a ( Jy) Fi 0 or I a (Jy) F z a ( Jy) F3:6 m ( Jy) F4:5 m ( Jy) F5:8 m ( Jy) F8:0 m (Jy) F24 m ( Jy) F70 m (mjy) F160 m (mjy) Reference SWIRE... LH_01 SWIRE J104148.93+592233.0 <0.441 <0.326 <0.486 <0.821... 5.4 12 <43 120 1351 <17.0 <90 1 LH_ 02 SWIRE J104605.56+583742.4 <0.441 <0.326 <0.486 <0.821... 12 20 69 124 1407 <17.0 <90 1 LH_A4 SWIRE J104409.95+585224.8 0.360 1.038 1.186 1.312 1.823 64 150 391 1093 4134 <17.0 <90 2 LH_A5 SWIRE J104453.07+585453.1 0.236 0.333 0.409 0.611 <1.000 69 141 292 467 1190 <17.0 <90 2 LH_A6 SWIRE J104613.48+585941.4 1.115 1.799 1.993 2.850 3.535 27 34 58 127 1147 <17.0 <90 2 LH_A8 SWIRE J104528.29+591326.7 <0.441 0.517 1.042 2.223 3.654 32 45 91 207 2462 9.8 <90 2 LH_A11 SWIRE J104314.93+585606.3 <0.411 <2.473 <2.324 <3.269... 9.2 22 60 117 968 <17.0 <90 2 N1_09 SWIRE J160532.69+535226.4 <0.801 1.107 1.491 1.914 <3.784 15 21 <43 73 2894 8.5 <90 1 N2_06 SWIRE J163511.43+412256.8 <0.801 <0.409 <0.752 <1.324 <3.784 14 18 36 116 4231 10.3 <90 1 N2_08 SWIRE J164216.93+410127.8 b <0.801 <0.409 <0.752 <1.324 <3.784 72 176 475 1178 3984 <17.0 80.5 1 N2_09 SWIRE J164401.40+405715.0 <0.801 <0.409 <0.752 <1.324 <3.784 39 58 86 143 3715 27.5 121 1 NDWFS... B1 SST24 J143001.91+334538.4 c... 0.142 0.302 0.521... 8 29 102 586 3800 <30.0... 1 B2 SST24 J143644.22+350627.4 d... 0.325 0.525 0.753... 43 97 380 732 2300 <30.0... 1 B3 SST24 J142827.19+354127.7... 0.980 2.290 5.210... 296 621 1601 3939 10550...... 1 B4 SST24 J142648.90+332927.2 e... 0.514 0.912 1.194... 53 178 493 911 2300 <30.0... 1 B5 SST24 J143508.49+334739.8... 0.390 0.692 0.905... 12 21 34 237 2600 <30.0... 1 FLS... MIPS 42 SST24 J171758.44+592816.8...... 0.186...... <9 30 103 680 4712 12.8... 3 MIPS 78 SST24 J171538.18+592540.1...... 0.186...... <9 39 <72 268 2973 <3.9... 3 MIPS 15840 SST24 J171922.40+600500.4...... 0.462...... 18 25 62 197 1821 <4.8... 3 MIPS 22204 SST24 J171844.38+592000.5...... 2.996...... <27 39 132 478 4101 13.2... 3 MIPS 22303 SST24 J171848.80+585115.1...... 0.395...... <9 <12 <54 103 2030 7.2... 3 a b c d e Notes. Typical uncertainties to the optical and IR fluxes are 4% and 10% of the measured fluxes, respectively. Upper limits correspond to 5. The optical data correspond to Ug 0 r 0 i 0 z in SWIRE, to UBRI in NDWFS, and to R in FLS. Also known as ELAISC15 J164216.9+410128. Source 9 in Houck et al. (2005). Also known as NDWFS J143644.3+350627 and source 13 in Houck et al. (2005). Also known as FIRST J142648.9+332927 and ELAIS R142650+332940B. References. (1) This work; (2) Weedman et al. 2006a; (3) IRAC data from Sajina et al. 2007a and R-band and MIPS[24] data from Yan et al. 2007.

964 POLLETTA ET AL. Vol. 675 TABLE 2 IRS Low-Resolution Spectroscopy Observation Log Source Name Program ID SL1 (s) LL2 (s) LL1 (s) LH_01... 136 3 ; 60 6 ; 120 6 ; 120 LH_02... 136 3 ; 60 6 ; 120 6 ; 120 LH_A4 a... 15 2 ; 60 3 ; 120 3 ; 120 LH_A5 a... 15 3 ; 60 6 ; 120 6 ; 120 LH_A6 a... 15 3 ; 60 6 ; 120 6 ; 120 LH_A8 a... 15 2 ; 60 3 ; 120 3 ; 120 LH_A11 a... 15 5 ; 60 6 ; 120 6 ; 120 N1_09... 15 3 ; 60 3 ; 120 3 ; 120 N2_06... 15 2 ; 60 2 ; 120 2 ; 120 N2_08... 15 2 ; 60 2 ; 120 2 ; 120 N2_09... 15 2 ; 60 2 ; 120 2 ; 120 B1... 12 1 ; 240 2 ; 120 2 ; 120 B2... 15 3 ; 60 3 ; 120 3 ; 120 B3... 15 2 ; 60 2 ; 120 2 ; 120 B4... 15 2 ; 60 3 ; 120 3 ; 120 B5... 15 3 ; 60 3 ; 120 3 ; 120 MIPS 42... 3748 2 ; 240 2 ; 120 3 ; 120 MIPS 78... 3748 2 ; 240 2 ; 120 3 ; 120 MIPS 15840... 3748 2 ; 240 6 ; 120 7 ; 120 MIPS 22204... 3748 2 ; 240 2 ; 120 3 ; 120 MIPS 22303... 3748... 2 ; 120 3 ; 120 Note. IRS data processed with ver. 13.0 of the SSC pipeline. a IRS data processed with ver. 11.0 of the SSC pipeline. LL1). The MIPS 24 m fluxes typically agree within 10% of the 24 m IRS flux. 4. SPECTRAL PROPERTIES OF THE IRS SPECTRA The IRS spectra of all 21 sources are shown in Figure 2. The spectra of an absorbed Seyfert 1 (Mrk 231; Weedman et al. 2005) and a heavily obscured ULIRG (IRAS F00183 7111, hereafter I00183; Tran et al. 2001; Spoon et al. 2004) are also shown for comparison. Both reference spectra exhibit the Si absorption feature at 9.7 m but with different optical depths, Si (Mrk 231) ¼ 0:65 (Spoon et al. 2002) and Si (I00183) 5 (Spoon et al. 2004). The spectrum of I00183 also shows absorption features at 4.26 and 4.67 m due to CO 2 and CO gas and at 6.85 and 7.25 m due to hydrogenated amorphous carbons ( HACs; Spoon et al. 2004). Spectroscopic redshifts are derived from the IRS spectra based on the location of the Si feature in comparison with the spectra of Mrk 231 and I00183. Only in two cases (LH_A4 and B3) are optical spectroscopic redshifts available ( Polletta et al. 2006; Desai et al. 2006). For most of the sources, IR spectroscopic redshifts were previously reported (Weedman et al. 2006a; Yan et al. 2007; Houck et al. 2005), although we have revised some of them. The estimated redshifts are reported in Table 3. Based on the comparison with previous z determinations and with the spectroscopic redshifts, we estimate that uncertainties associated with these redshifts are as much as 0.2 for sources with the poorest S/N. The Si feature in absorption is clearly observed in 18 objects (LH_02, LH_A4, LH_A5, LH_A6, LH_A8, LH_A11, N1_09, N2_08, N2_09, B1, B2, B3, B4, B5, MIPS 42, MIPS 78, MIPS 22204, MIPS 22303) and only marginally observed in three objects ( LH_01, N2_06, MIPS 15840). The apparent optical depth of the Si feature, Si,ismeasuredasln(F9:7 int obs m /F9:7 m ), where F9:7 obs m is the observed flux at 9.7 m, at the minimum of the Si feature, and F9:7 int m is the intrinsic flux at 9.7 m that would arise from an extrapolated, unabsorbed continuum. Since in most of the cases the SEDs and spectra do not sample the wavelength region beyond the Si feature, we cannot use the standard method to derive the intrinsic continuum (see, e.g., Spoon et al. 2004). The intrinsic flux at 9.7 m is thus estimated using two different methods, (1) by extrapolating a 4Y8 m rest-frame power-law fit to 9.7 m, and (2) by normalizing an unobscured AGN template to the observed 7.4 m flux, after redshifting it to the redshift of each source, and reddening it using the Galactic center extinction curve (Chiar et al. 2006) until the observed Si feature is well reproduced. The Si optical depth, Si, is measured only in sources at z < 2:7 (18 sources). For the remaining we estimate a lower limit since we have only an upper limit to the flux at the bottom of the Si feature. The measured values, reported in Table 3, vary from >0.2 to 3.4. The expected locations of other spectral features common in starburst galaxies and AGNs are also annotated in Figure 2: (1) absorption features due to molecular gas, CO 2, at 4.26 and 4.67 m, to water ice at 6.15 and 7.67 m, and to HACs at 6.85 and 7.25 m; and (2) emission features associated with polycyclic aromatic hydrocarbons (PAHs) at k ¼ 3:3, 6.22, 7.7, 8.6, and 11.3 m, and with low- and high-ionization emission lines, [Ar ii] 6.99m, H 2 S(3) 9.66 m, and [S iv] 10.54m. Absorption features like CO and HACs would imply the presence of warm gas along the LOS, and features like CO 2, water ice, and Si would imply the presence of shielded cold molecular clouds (Spoon et al. 2004, 2007). Although there are sometimes features in the individual spectra that are consistent with these spectral features, higher S/N spectra would be necessary to confirm their detection. No spectrum shows any secure feature except the Si absorption. 5. MODELING THE SPECTRAL ENERGY DISTRIBUTIONS WITH TORUS MODELS The opticalyir SEDs and rebinned IRS spectra of all 21 sources are shown in Figure 1. The IR SEDs, from 3.6 to 24 m, and IRS spectra are modeled with a grid of torus models from Hönig et al. (2006), and the residuals in the NIR are fitted with a galaxy template to represent the host galaxy. The model parameters are the cloud density distribution (from compact to extended), the vertical radial distribution (with various degrees of flaring or nonflaring), the number of clouds along the LOS (optical depth), and the torus inclination (from face-on to edge-on). The best model is the one that gives the best fit, based on a 2 test, to the five broadband photometric data points from 3.6 to 24 m and to a maximum of six additional data points derived by interpolating the IRS spectrum at k ¼ 3, 5, 7, 8.5, 9.7, and 12 m in the rest frame. The best-fit models are shown in Figure 1, and the best-fit parameters are listed in Table 4. In order to fit the stellar component, we adopt a 3 Gyr old elliptical template (from GRASIL; Silva et al.1998) normalized at the observed flux of the residuals (modelsubtracted SED) at 3.6 m (in the Z band for LH_A5). The choice of an elliptical template is justified by the evidence of ellipticals as hosts in the vast majority of quasars (e.g., Dunlop et al. 2003). In order to check the validity of our assumption on the host galaxy, we estimate the associated R-band (k rest ¼ 0:7 m) absolute magnitudes, M R, and compare them with those measured for the large sample of quasars in Dunlop et al. (2003). The measured M R range from 25.9 to 22.2, and the median value is 23.7. These values are consistent with those found by Dunlop et al. (2003), M R ¼ 23:53 0:09, but our sample shows a wider dispersion that can be attributed to the large uncertainty of our method and to the different redshift range. A host galaxy of later type cannot be ruled out, but the low number of detections in the optical, the lack of NIR (JHK ) data, and contamination from the AGN light do not allow us to better constrain the host type. In most of the cases, the elliptical template provides an acceptable fit.

No. 2, 2008 OBSCURATION IN EXTREMELY LUMINOUS QUASARS 965 Fig. 1. Rest-frame SEDs in L vs. k (black symbols and line) and best-fit model (solid gray line) obtained from a torus model (dashed gray line) and an elliptical template (dotted gray line) to fit the residuals in the NIR. Open diamonds represent ISO 15 m (in N2_08) and MAMBO 1.2 mm (in all E-FLS sources) data (Gonzalez- Solares et al. 2005; Lutz et al. 2005). Downward-pointing arrows represent 5 upper limits for the optical and infrared data points and 3 upper limits for millimeter data. Source names and redshifts are annotated. Only in one source, LH_A8, is the SED of the stellar component not well fitted with an elliptical template because of an excess of emission at k rest < 0:3 m and could a late spiral template provide a better fit. Note that a heavily obscured starburst galaxy would have an opticalynir SED that is very similar to that of an elliptical galaxy. The reduced 2 obtained by comparing the model with the data at k > 1 m in the rest frame are reported in Table 4. We refer to these as T (for torus) models (see also Table 4). In 11 cases, a T model well reproduces the observed IR SED and spectrum ( 2 < 1). In the remaining 10 cases, poorer fits are obtained (1 < 2 < 10). In three cases the model underestimates the NIR emission (LH_A4, LH_A5, and N2_08), in four cases the model fails to reproduce the depth of the Si feature (B1, B4, MPS78, and MIPS 22303), and in the remaining three cases the large 2 values are caused by either a noisy IRS spectrum (LH_02 and N2_09) or a poor fit at short wavelengths ( N2_09 and MIPS 22304). There are also two sources with good 2 but their fits do not well reproduce the Si feature. This is due to the fact that the feature is not well sampled by the representative values for the IRS spectrum. There is indeed only one data point in the Si feature. The main cause of failure in the fits with large 2 is the inability of the models to reproduce both a prominent NIR emission and a deep Si absorption feature. Yet, sources with deep Si features are expected to have weak NIR emission (e.g., Pier & Krolik 1992; Levenson et al. 2006). An NIR excess requires that hot dust is seen directly by the observer (typically happening only in less obscured objects), while the Si absorption feature requires

966 POLLETTA ET AL. Vol. 675 Fig. 1 Continued the presence of a significant amount of optically and geometrically thick dust (e.g., Pier & Krolik 1992; Granato et al. 1997; Levenson et al. 2006; Imanishi et al. 2007), and this cooler dust component should absorb the hot dust emission from the vicinity of the obscured AGN. We attempted, therefore, to model the objects with poor fits, as well as the rest of the sample with an alternative model that includes a cold absorber detached from the torus ( hereafter T+C models). We assume the Galactic center extinction curve (Chiar et al. 2006) for the cold absorber. We do not model reemission from this absorber for simplicity and because this is expected to occur at far-infrared ( FIR) wavelengths. The number of degrees of freedom in the 2 estimates is 4 in the case of T models and 5 in the case of T+C models. The cold absorber is able to produce deeper Si features and significantly improves the fits for nine sources, as shown in Figure 3. In two cases, for B2 and B5, the 2 have actually increased, but they are still <1. In these two sources the 2 is more sensitive to the broadband photometric data than to the IRS spectrum because of our limited sampling in the observed deep Si feature. But the two fits are equally good, and from a visual inspection of the fits, we choose the model with the cold absorber component as the preferred model because it better fits the Si feature. The T+C models were adopted for these two sources and for seven sources for which the T+C model gives a significantly better fit (LH_A4, LH_A5, N2_08, B1, B2, B4, B5, MIPS 78, and MIPS 22303). In two cases (B4 and N2_08), even the T+C model does not provide a good fit to the NIR emission, although it is an improvement with respect to the T model. No improvement was obtained for the three sources with poor 2 (LH _02, N2_09, and MIPS 22204); therefore, for these we keep the simple T model. In addition to the T+C models, we also show in Figure 3 the torus model before applying the extinction due to the cold absorber (see the dot-dashed line in Fig. 3). The cold absorber can absorb up to 84% of torus emission in the NIR and MIR (1Y50 m), as shown in Figure 3. In modeling the SEDs, we neglected the optical data. However, we notice that in most of the cases the predicted optical flux from the torus model is lower than the observed optical emission. The optical light might be associated with the host galaxy or with scattered light from the AGN. In six sources detected in the optical, the SED shows an upturn or a blue continuum toward shorter wavelengths, i.e., in the rest-frame far-ultraviolet. This feature can be reproduced by either a population of young stars or emission from the AGN accretion disk. The latter scenario is favored by the optical spectrum of source LH_A4 (aka SW 104409). Its optical spectrum is dominated by a blue faint continuum and narrow emission lines, e.g., Ly and C iv k1549 with asymmetric and weak broad components ( Polletta et al. 2006). The most likely explanation for the properties of the observed optical spectrum is that it is dominated by scattered light, and the scattering fraction is estimated to be <1% (Polletta et al. 2006). In one case (N2_06), the LOS is at only 30 from the torus axis, implying an almost clear view of the nuclear region. Since no

No. 2, 2008 OBSCURATION IN EXTREMELY LUMINOUS QUASARS 967 Fig. 2. Observed IRS spectra (gray solid line) and broadband IR photometric data (black filled circles) of all selected sources. The expected locations of some spectral features are annotated at the corresponding observed wavelength (P refers to PAH emission features). The IRS spectra of the Seyfert 1 Mrk 231 (dotted line) and of the heavily obscured AGN and ULIRG IRAS 000183 7111 (dashed line) are overplotted for comparison. The name and redshift of each source are annotated, and rest-frame wavelengths are reported on the upper horizontal axis. additional absorber is required by the data, we expect to see the nuclear emission directly. This seems in apparent contradiction with the faint optical emission of this object. Its best-fit model, however, predicts one optically thick cloud, associated with the torus, along the LOS (see Table 4). The presence of such a cloud in front of the optical source would be enough to absorb the intrinsic optical emission in this source. The estimated apparent Si optical depth in the model is Si ¼ 0:07Y0:34, which corresponds to a visual optical depth of 0.4Y2.1ortoasuppression factor of the optical flux of 40%Y0.8%. However, Si is poorly constrained by the data and in the model because this source is at z ¼ 2:96. The red MIR SED of this object would imply a much higher optical depth than derived by the Si feature. Because of its high z, the dust distribution in N2_06 is thus poorly constrained. A deeper Si might be present, implying a higher torus inclination or the presence of a cold absorber. On the other hand, it is also possible that our best-fit model is correct and that the source has only a weak Si absorption feature and a low-inclination torus, and one optically thick cloud is responsible for suppressing the optical and NIR nuclear light. This scenario might be quite common in obscured QSOs as suggested by a recent study of the IRS spectra of a sample of type 2 absorbed QSOs (Sturm et al. 2006). 5.1. Model Parameters Here we analyze the model parameters of the best-fit models. In the case of the nine sources for which the T+C model is preferred, we consider only the parameters obtained with such a model (see Table 4). The cloud density distribution, which can be considered as an indicator for a compact or extended torus, is expressed by a power law, n r (r) / r a,wherer is the torus radius. In the models grid, the index a varies from 1 to 3 in steps of 0.5. Larger values of the index a indicate more compact distributions. Since values lower than 1 are not supported by theoretical considerations for an accretion scenario (Beckert & Duschl 2004), we set a minimum value for the parameter a of unity. In models with a 1, the depth of the Si feature increases for larger values of a (see Fig. 10 in Hönig et al. 2006). Trends of Si strength with a are different for our models based on an accretion torus compared to the models in Levenson et al. (2007). They do not model the dust distribution in a torus but

968 POLLETTA ET AL. Vol. 675 Fig. 2 Continued instead use different dust distributions (slab and shell) around a central heating source, and they do not consider the combination of a clumpy medium and an additional absorber as proposed here. According to Levenson et al. (2007), the strength of the Si absorption feature is a function of the temperature contrast in dust material, and, thus, a deep absorption feature favors a more extended dust distribution, i.e., a ¼ 0Y1. However, for the majority of our sources, a more compact dust distribution is favored; a ¼ 3 is chosen in nine cases, a ¼ 2 in 10 cases, and smaller values, a 1:5, are preferred in only two sources. The clouds vertical distribution (flaring or nonflaring) is approximated by a power law, H(r) / r b, where H is the torus scale height and r is the torus radius. In the models grid, the index b can assume values 1, 1.5, and 2. A nonflaring distribution (b ¼ 1)ispreferredby17sources, and moderately flaring (b ¼ 1:5) is preferred only by four sources. The torus inclination is defined by the angle between the torus axis and the LOS. A face-on torus has ¼ 0, and an edgeon torus has ¼ 90. In the models grid, the angle varies from 0to90 in steps of 15. Our sources do not show a preferred torus inclination; indeed, ranges from 15 to 90, with most of the sources (13) at intermediate values ( ¼ 30 Y45 ). All but two (B5 and MIPS 22303) of the sources modeled with the T+C model and one of those with weak Si absorption feature (N2_06) favor a torus with little inclination, ¼ 0 Y30. All of the others have inclined tori, 45. Therefore, optical obscuration or Si in absorption does not necessarily imply an LOS intercepting the torus (see also Rigby et al. 2006; Brand et al. 2007). An additional model parameter is the number of clouds along the LOS, N0 LOS. This parameter is the average number of clouds obtained from five different cloud arrangements for the same set of model parameters and is thus indicative of the extinction to the AGN emission produced by the torus, V T N 0 LOS (Natta & Panagia 1984). The apparent optical depths in the Si feature of the Tand T+C models, Si T TþC, and Si, are measured following the same procedure applied to the data (see x 4). The apparent optical depths so derived are reported in Table 4. The LOS optical depths associated with the torus in the T models are Si T 1:04, with a median value of Si T ¼ 0:13 or 0.46 depending on the method appliedtoestimatesi T. The optical depths associated with the cold absorber are V C ¼ 4Y25, with a median value of 15. The apparent optical depths for the sources fitted with the T+C model, Si TþC, range from 0.6 to 2.9 with a median value of 1.6 or from 0.4 to 1.4 with a median value of 0.9 depending on the method used to measure it. In Figure 4 we compare the modeled optical depths in the Si feature, Si mod, with the observed optical depths, Si meas (see x 4). The modeled optical depths range from 0.01 to 2.9, considering

No. 2, 2008 OBSCURATION IN EXTREMELY LUMINOUS QUASARS 969 Fig. 2 Continued both the T and T+C models. The modeled and measured Si satisfy the following relation: Si mod ¼ (0:96 0:13)Si meas 0:03 0:13. Although there is a general agreement between the two estimates, it is clear that these kinds of measurements are characterized by large uncertainties, especially in objects at high redshift for which the Si feature is not well sampled, and that it is a difficult task to model both the NIRYMIR continuum and the Si absorption feature. In summary, we find that there is no preferred torus inclination associated with the detection of a Si feature in absorption. However, in the cases where the absorption feature is well detected and can be modeled by the torus model, the inclination is always higher than ¼ 45, consistent with the LOS intercepting the torus. Sources with a less inclined torus require an additional absorber to explain the observed Si feature or show a weak Si feature. A compact nonflaring torus is preferred by the majority of the sources. The clouds radial density distribution indeed indicates that the torus emission region is compact, or that the NIR and MIR emission is dominated by dust in the vicinity of the nucleus. The preferred nonflaring torus in the majority of these luminous sources is consistent with the predictions of the receding torus models (e.g., Simpson 2005; Hönig & Beckert 2007). According to these models, the opening angle of the torus increases at larger luminosities. Our results are in agreement with the predictions from Hönig & Beckert (2007). These authors claim that flaring should not occur in high-luminosity sources because large clouds at large distances from the AGN should be driven away by the radiation pressure. 5.2. Far-IR Emission According to current AGN evolutionary models, obscured and extremely luminous AGNs are believed to represent a specific and rare phase in the evolution of an AGN, when the central supermassive black hole (SMBH) is fully grown and still surrounded by a large amount of dust and gas (Sanders et al.1988; Di Matteo et al. 2005; Hopkins et al. 2005). High-luminosity AGNs are believed to be triggered by large-scale galaxy mergers and, therefore, to be accompanied by intense starburst activity. In order to test whether this scenario applies to our sample, we search for starburst signatures in our objects. Obscured starburst galaxies are generally heavily extincted at optical wavelengths and are weak X-ray sources compared to AGNs. Thus, the only observations that could reveal a starburst are PAH features at MIR wavelengths and strong continuum from cool dust at FIR and submillimeter

970 POLLETTA ET AL. Vol. 675 Fig. 2 Continued wavelengths. An analysis of the starburst contribution based on the PAH features is discussed in x 9.1; here we analyze the FIR properties. Most of the power produced by a starburst emerges in the FIR. The typical FIR emission of powerful starbursts is characterized by luminosities 10 11.0 Y10 12.5 L and peaks at 60Y100 m (Sanders & Mirabel 1996). Such luminosities correspond to star formation rates (SFRs) of 20Y550 M yr 1 (Kennicutt 1998). Higher rates, up to 5000 M yr 1, and luminosities, up to 3 ; 10 13 L, are measured in some high-z ULIRGs/submillimeter galaxies (SMGs) and attributed to starbursts (e.g., Chapman et al. 2004). In our sample, five SWIRE sources are detected at 70 or 160 m and three E-FLS sources are detected at 70 m. Note that both MIPS FIR observations are available only for SWIRE sources (13 sources), and MIPS 70 m pointed observations are available for the E-FLS sources (Sajina et al. 2007a). The SWIRE observations in the FIR are sensitive enough only to reveal FIR luminosities >10 13 L at the observed redshifts (the 5 limits are 18 and 108 mjy at 70 and 160 m, respectively). In Figure 5 we compare the SWIRE 5 detection limit and the detected fluxes at 70 and 160 m of our sources with those expected for two starburst galaxies, M82 and Arp 220, with an IR luminosity of 10 11 and 10 12.5 L at 1 < z < 3:2. The figure clearly shows that we do not expect to detect the FIR emission of starbursts with such luminosities at z > 1:2. What is then the origin of the detected FIR emission in our objects? We investigate three possible origins: the torus, AGNheated dust at large distances from the nucleus, and an exceptionally powerful starburst. The spectrum of our torus models peaks at around 10 m, corresponding to a temperature of about 300 K, and falls off at longer wavelengths. In two cases, the 70 m detections or upper limits agree well with our models (N1_09 and N2_06), but in the other six cases, the observed 70 or 160 m fluxes are significantly higher than the model predictions (LH_A8, N2_08, N2_09, MIPS 42, MIPS 22204, and MIPS 22303) and require an additional component to be explained. The SEDs of these sources resemble that of the starburst /AGN composite source CXO J1417 discussed in Le Floc h et al. (2007). The FIR luminosity and SFR estimated for CXO J1417 are 4:5 ; 10 12 L and 750 M yr 1, respectively, as in a powerful starburst. In order to quantify the amount of luminosity detected in the FIR, we model the observed fluxes at rest frame k > 5 mwithastarburst template. We first derive the residual fluxes after subtracting

No. 2, 2008 OBSCURATION IN EXTREMELY LUMINOUS QUASARS 971 TABLE 3 Infrared Luminosities and Optical Depths Source ID z L 5:5 m L 5:8 m L 6:0 m L AGN a IR L AGN b bol L SB c FIR L AGNþSB d FIR L AGNþSB e IR L AGNþSB f bol fir AGN g fir SB h Si PL i Temp j Si SFR k LH_01... 2.84 46.07 46.09 46.09 46.39 46.45.................. >0.79 >0.37... LH_02... 2.12 45.63 45.66 45.67 45.95 46.00.................. 1.98 0.91... LH_A4... 2.54 l 46.45 46.45 46.45 46.69 46.83.................. 0.87 0.57... LH_A5... 1.94 m 45.65 45.65 45.64 45.89 46.08.................. 0.63 0.41... LH_A6... 2.20 m 45.67 45.68 45.68 45.92 46.02.................. 0.43 0.18... LH_A8... 2.42 m 46.12 46.15 46.16 46.42 46.49 46.34 46.36 46.83 46.86 0.39 0.61 0.86 0.33 994 LH_A11... 2.30 m 45.70 45.72 45.73 46.01 46.08.................. 0.85 0.38... N1_09... 2.75 46.35 46.38 46.40 46.69 46.72 46.10 46.19 46.86 46.88 0.67 0.33 >0.86 >0.42 572 N2_06... 2.96 46.60 46.61 46.61 46.89 46.92 46.16 46.20 47.02 47.04 0.74 0.26 >0.67 >0.20 657 N2_08... 2.40 46.42 46.42 46.42 46.59 46.69 46.80 46.81 47.20 47.23 0.25 0.75 1.78 1.14 2868 N2_09... 1.98 46.05 46.08 46.09 46.41 46.46 46.75 46.76 47.12 47.14 0.20 0.80 1.01 0.15 2556 B1... 2.46 46.41 46.41 46.41 46.56 46.62.................. 2.36 1.49... B2... 1.77 45.91 45.91 45.90 46.09 46.20.................. 0.88 0.71... B3... 1.293 l 46.29 46.30 46.30 46.53 46.60.................. 0.43 0.21... B4... 1.82 45.78 45.78 45.78 45.98 46.11.................. 1.72 1.21... B5... 2.00 46.01 46.03 46.04 46.26 46.30.................. 2.43 1.39... MIPS 42... 1.90 n 46.20 46.21 46.21 46.44 46.48 46.13 46.16 46.72 46.75 0.52 0.48 0.42 0.13 613 MIPS 78... 2.43 n 46.21 46.22 46.23 46.52 46.58.................. 1.04 0.58... MIPS 15840... 2.23 n 45.96 45.97 45.97 46.21 46.28.................. 0.37 0.10... MIPS 22204... 2.08 46.17 46.19 46.20 46.48 46.55 46.18 46.22 46.77 46.81 0.52 0.48 1.04 0.58 688 MIPS 22303... 2.50 n 46.11 46.13 46.14 46.28 46.31 46.34 46.36 46.78 46.80 0.31 0.69 1.10 0.80 994 Notes. All luminosities are logarithm in units of ergs s 1. Luminosities for a starburst component are estimated only for sources detected at 70 or 160 m. a Logarithm of the AGN IR luminosity obtained by integrating the torus model between 3 and 1000 m. b Logarithm of the AGN bolometric luminosity obtained by integrating the torus model between 0.1 and 1000 m. c Logarithm of the starburst component FIR luminosity between 42.5 and 122.5 m. d Logarithm of the AGN and starburst FIR luminosities between 42.5 and 122.5 m. e Logarithm of the AGN and starburst IR luminosities between 3 and 1000 m. f Logarithm of the AGN and starburst bolometric luminosities. g Fraction of AGN luminosity to the total IR luminosity. h Fraction of starburst luminosity to the total IR luminosity. i Silicate optical depth derived from ln (F int obs int /F ) with F estimated extrapolating a power-law model fit to the data at k < 7 m in the rest frame. j Silicate optical depth derived from ln (F int of the source. k Star formation rate derived from L SB 9:7 m 9:7 m 9:7 m /F obs 9:7 m )withf int 9:7 m 9:7 m estimated from a type 1 QSO template normalized at the observed 24 m flux and redshifted at the redshift FIR using the Kennicutt (1998) relationship. l Optical spectroscopic redshift for LH_A4 from Polletta et al. (2006) and for B3 from Desai et al. (2006). m LH_A5: z ¼ 1:89; LH_A6: z ¼ 2:10; LH_A8: z ¼ 2:31; LH_A11: z ¼ 2:25; in Weedman et al. (2006a). n MIPS 42: z ¼ 1:95 0:07; MIPS 78: z ¼ 2:65 0:1; MIPS 15840: z ¼ 2:3 0:1; MIPS 22303: z ¼ 2:34 0:14; in Yan et al. (2007). the torus contribution to the observed FIR fluxes and then fit them with starburst templates. Two different starburst templates, M82 and Arp 220 (Silva et al.1998), are used for the fits, but we only show the results obtained with the M82 template because it provides better fits in the majority of the cases and similar good fits in the remaining cases. The starburst fits, combined with the torus and host galaxy fits, are shown in Figure 6. The estimated starburst luminosities and contribution to the total IR luminosity are reported in Table 3. In half of the FIR-detected sources the torus contribution to the total IR luminosity is lower than that of the additional FIR component. The remaining objects divide equally between those for which the two contributions are similar and those for which the torus contribution is larger. The estimated starburst FIR luminosities are all greater than 3:3 ; 10 12 L.These luminosities imply SFRs >600 M yr 1 (Kennicutt1998), which are consistent only with those measured in the most powerful starbursts. Another possible origin of the FIR component is reemission of the energy absorbed by the cold absorber discussed in x 5. There are two sources detected at long wavelengths and modeled with the T+C model, N2_08 and MIPS 22303. The estimated absorbed luminosities in the 1Y50 m wavelength range are 1:8 ; 10 47 and 5:25 ; 10 45 ergs s 1 for N2_08 and MIPS 22303, respectively. These correspond to 2.8 and 0.24 times the FIR luminosities derived from the starburst template (LFIR SB ). Although these ratios depend on the cold absorber covering factor and on the efficiency in reemitting the absorbed energy, these values indicate that this hypothesis is energetically viable. Although we can reproduce the FIR component with a starburst template or with reemission from the cold absorber, its origin remains undetermined. Large FIR luminosities, as measured in our sources, even when cold molecular gas is detected, do not probe the presence of a powerful starburst, e.g., in QSOs where [O ii] k3727 emission is not detected (Ho 2005). Thus, both explanations remain plausible and do not exclude others. In order to understand the origin of the FIR component, we would need more FIR measurements, higher S/N IR spectra, and NIR spectra. Multiple FIR detections would help to constrain the radiation field, high-s/n IR spectra would allow us to search for and, eventually, measure PAH emission (see x 9.1), and NIR spectra could be used to constrain the SFR from the [O ii] k3727 emission line. 5.3. AGN Bolometric Luminosity Assuming the best-fit torus model and integrating it from 1000 8 to 1000 m, we can derive a lower limit to the AGN bolometric luminosity. This estimate of AGN bolometric luminosity does not include the AGN optical light that is not reprocessed by the